graphwise 1.1.1 → 1.2.0

package/dist/utils/index.cjs

@@ -1,13 +1,176 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
-const require_kmeans = require("../kmeans-B0HEOU6k.cjs");
-const require_utils = require("../utils-spZa1ZvS.cjs");
-exports.approximateClusteringCoefficient = require_utils.approximateClusteringCoefficient;
-exports.batchClusteringCoefficients = require_utils.batchClusteringCoefficients;
-exports.entropyFromCounts = require_utils.entropyFromCounts;
-exports.localClusteringCoefficient = require_utils.localClusteringCoefficient;
-exports.localTypeEntropy = require_utils.localTypeEntropy;
+const require_kmeans = require("../kmeans-BIgSyGKu.cjs");
+//#region src/utils/clustering-coefficient.ts
+/**
+* Compute the local clustering coefficient for a single node.
+*
+* The clustering coefficient is defined as:
+*   CC(v) = (triangles through v) / (possible triangles)
+*   CC(v) = 2 * |{(u,w) : u,w in N(v), (u,w) in E}| / (deg(v) * (deg(v) - 1))
+*
+* For nodes with degree < 2, the clustering coefficient is 0.
+*
+* @param graph - The graph to compute on
+* @param nodeId - The node to compute clustering coefficient for
+* @returns The clustering coefficient in [0, 1], or 0 if undefined
+*/
+function localClusteringCoefficient(graph, nodeId) {
+	const neighbours = [...graph.neighbours(nodeId, "both")];
+	const degree = neighbours.length;
+	if (degree < 2) return 0;
+	let triangleCount = 0;
+	for (let i = 0; i < neighbours.length; i++) {
+		const u = neighbours[i];
+		if (u === void 0) continue;
+		for (let j = i + 1; j < neighbours.length; j++) {
+			const w = neighbours[j];
+			if (w === void 0) continue;
+			if (graph.getEdge(u, w) !== void 0 || graph.getEdge(w, u) !== void 0) triangleCount++;
+		}
+	}
+	const possibleTriangles = degree * (degree - 1) / 2;
+	return triangleCount / possibleTriangles;
+}
+/**
+* Compute approximate local clustering coefficient using sampling.
+*
+* For nodes with many neighbours, this samples neighbour pairs rather than
+* checking all pairs. Useful for large graphs where exact computation is expensive.
+*
+* @param graph - The graph to compute on
+* @param nodeId - The node to compute clustering coefficient for
+* @param sampleSize - Maximum number of neighbour pairs to check (default: 100)
+* @returns The approximate clustering coefficient in [0, 1]
+*/
+function approximateClusteringCoefficient(graph, nodeId, sampleSize = 100) {
+	const neighbours = [...graph.neighbours(nodeId, "both")];
+	const degree = neighbours.length;
+	if (degree < 2) return 0;
+	const possibleTriangles = degree * (degree - 1) / 2;
+	if (possibleTriangles <= sampleSize) return localClusteringCoefficient(graph, nodeId);
+	let triangleCount = 0;
+	let sampled = 0;
+	for (let i = 0; i < neighbours.length && sampled < sampleSize; i++) {
+		const u = neighbours[i];
+		if (u === void 0) continue;
+		for (let j = i + 1; j < neighbours.length && sampled < sampleSize; j++) {
+			const w = neighbours[j];
+			if (w === void 0) continue;
+			sampled++;
+			if (graph.getEdge(u, w) !== void 0 || graph.getEdge(w, u) !== void 0) triangleCount++;
+		}
+	}
+	return triangleCount / sampled * (possibleTriangles / possibleTriangles);
+}
+/**
+* Compute clustering coefficients for multiple nodes efficiently.
+*
+* Reuses neighbour sets to avoid repeated iteration.
+*
+* @param graph - The graph to compute on
+* @param nodeIds - The nodes to compute clustering coefficients for
+* @returns Map from nodeId to clustering coefficient
+*/
+function batchClusteringCoefficients(graph, nodeIds) {
+	const results = /* @__PURE__ */ new Map();
+	for (const nodeId of nodeIds) results.set(nodeId, localClusteringCoefficient(graph, nodeId));
+	return results;
+}
+//#endregion
+//#region src/utils/entropy.ts
+/**
+* Entropy computation utilities for graph analysis.
+*
+* Shannon entropy measures uncertainty or randomness in a distribution.
+* Used in EDGE and HAE algorithms for heterogeneity-aware expansion.
+*
+* @packageDocumentation
+*/
+/**
+* Compute Shannon entropy of a probability distribution.
+*
+* Shannon entropy is defined as:
+*   H(X) = -Σ p(x) × log₂(p(x))
+*
+* A uniform distribution has maximum entropy.
+* A deterministic distribution (all probability on one value) has zero entropy.
+*
+* @param probabilities - Array of probabilities (should sum to 1)
+* @returns Entropy in bits (log base 2), or 0 if probabilities are invalid
+*/
+function shannonEntropy(probabilities) {
+	if (probabilities.length === 0) return 0;
+	let entropy = 0;
+	for (const p of probabilities) if (p > 0) entropy -= p * Math.log2(p);
+	return entropy;
+}
+/**
+* Compute normalised entropy (entropy divided by maximum possible entropy).
+*
+* Normalised entropy is in [0, 1], where:
+* - 0 means the distribution is deterministic (all mass on one value)
+* - 1 means the distribution is uniform (maximum uncertainty)
+*
+* This is useful for comparing entropy across distributions with different
+* numbers of possible values.
+*
+* @param probabilities - Array of probabilities (should sum to 1)
+* @returns Normalised entropy in [0, 1], or 0 if only one category
+*/
+function normalisedEntropy(probabilities) {
+	if (probabilities.length <= 1) return 0;
+	const H = shannonEntropy(probabilities);
+	const Hmax = Math.log2(probabilities.length);
+	if (Hmax === 0) return 0;
+	return H / Hmax;
+}
+/**
+* Compute entropy from a frequency count.
+*
+* Converts counts to probabilities and then computes entropy.
+* This is a convenience function when you have raw counts rather than
+* normalised probabilities.
+*
+* @param counts - Array of frequency counts
+* @returns Entropy in bits
+*/
+function entropyFromCounts(counts) {
+	if (counts.length === 0) return 0;
+	const total = counts.reduce((sum, c) => sum + c, 0);
+	if (total === 0) return 0;
+	return shannonEntropy(counts.map((c) => c / total));
+}
+/**
+* Compute local type entropy for a node's neighbours.
+*
+* This measures the diversity of types among a node's neighbours.
+* High entropy = heterogeneous neighbourhood (diverse types).
+* Low entropy = homogeneous neighbourhood (similar types).
+*
+* @param neighbourTypes - Array of type labels for neighbours
+* @returns Normalised entropy in [0, 1]
+*/
+function localTypeEntropy(neighbourTypes) {
+	if (neighbourTypes.length <= 1) return 0;
+	const typeCounts = /* @__PURE__ */ new Map();
+	for (const t of neighbourTypes) typeCounts.set(t, (typeCounts.get(t) ?? 0) + 1);
+	if (typeCounts.size === 1) return 0;
+	const probabilities = [];
+	const total = neighbourTypes.length;
+	for (const count of typeCounts.values()) probabilities.push(count / total);
+	return normalisedEntropy(probabilities);
+}
+//#endregion
+exports._computeMean = require_kmeans._computeMean;
+exports.approximateClusteringCoefficient = approximateClusteringCoefficient;
+exports.batchClusteringCoefficients = batchClusteringCoefficients;
+exports.entropyFromCounts = entropyFromCounts;
+exports.localClusteringCoefficient = localClusteringCoefficient;
+exports.localTypeEntropy = localTypeEntropy;
 exports.miniBatchKMeans = require_kmeans.miniBatchKMeans;
 exports.normaliseFeatures = require_kmeans.normaliseFeatures;
-exports.normalisedEntropy = require_utils.normalisedEntropy;
-exports.shannonEntropy = require_utils.shannonEntropy;
 exports.zScoreNormalise = require_kmeans.normaliseFeatures;
+exports.normalisedEntropy = normalisedEntropy;
+exports.shannonEntropy = shannonEntropy;
+
+//# sourceMappingURL=index.cjs.map

package/dist/utils/index.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.cjs","names":[],"sources":["../../src/utils/clustering-coefficient.ts","../../src/utils/entropy.ts"],"sourcesContent":["/**\n * Clustering coefficient computation for graph nodes.\n *\n * The local clustering coefficient measures how close a node's neighbours\n * are to being a complete graph (clique). It is used in SPAN MI variant\n * and GRASP seed selection.\n *\n * @packageDocumentation\n */\n\nimport type { ReadableGraph, NodeId } from \"../graph\";\n\n/**\n * Compute the local clustering coefficient for a single node.\n *\n * The clustering coefficient is defined as:\n *   CC(v) = (triangles through v) / (possible triangles)\n *   CC(v) = 2 * |{(u,w) : u,w in N(v), (u,w) in E}| / (deg(v) * (deg(v) - 1))\n *\n * For nodes with degree < 2, the clustering coefficient is 0.\n *\n * @param graph - The graph to compute on\n * @param nodeId - The node to compute clustering coefficient for\n * @returns The clustering coefficient in [0, 1], or 0 if undefined\n */\nexport function localClusteringCoefficient(\n\tgraph: ReadableGraph,\n\tnodeId: NodeId,\n): number {\n\tconst neighbours = [...graph.neighbours(nodeId, \"both\")];\n\tconst degree = neighbours.length;\n\n\t// Nodes with degree < 2 have no possible triangles\n\tif (degree < 2) {\n\t\treturn 0;\n\t}\n\n\t// Count actual triangles: pairs of neighbours that are connected\n\tlet triangleCount = 0;\n\n\tfor (let i = 0; i < neighbours.length; i++) {\n\t\tconst u = neighbours[i];\n\t\tif (u === undefined) continue;\n\n\t\tfor (let j = i + 1; j < neighbours.length; j++) {\n\t\t\tconst w = neighbours[j];\n\t\t\tif (w === undefined) continue;\n\n\t\t\t// Check if u and w are connected\n\t\t\tif (\n\t\t\t\tgraph.getEdge(u, w) !== undefined ||\n\t\t\t\tgraph.getEdge(w, u) !== undefined\n\t\t\t) {\n\t\t\t\ttriangleCount++;\n\t\t\t}\n\t\t}\n\t}\n\n\t// Possible triangles: deg * (deg - 1) / 2 pairs\n\t// We multiply by 2 because each triangle is counted once\n\tconst possibleTriangles = (degree * (degree - 1)) / 2;\n\n\treturn triangleCount / possibleTriangles;\n}\n\n/**\n * Compute approximate local clustering coefficient using sampling.\n *\n * For nodes with many neighbours, this samples neighbour pairs rather than\n * checking all pairs. Useful for large graphs where exact computation is expensive.\n *\n * @param graph - The graph to compute on\n * @param nodeId - The node to compute clustering coefficient for\n * @param sampleSize - Maximum number of neighbour pairs to check (default: 100)\n * @returns The approximate clustering coefficient in [0, 1]\n */\nexport function approximateClusteringCoefficient(\n\tgraph: ReadableGraph,\n\tnodeId: NodeId,\n\tsampleSize = 100,\n): number {\n\tconst neighbours = [...graph.neighbours(nodeId, \"both\")];\n\tconst degree = neighbours.length;\n\n\tif (degree < 2) {\n\t\treturn 0;\n\t}\n\n\tconst possibleTriangles = (degree * (degree - 1)) / 2;\n\n\t// If all pairs can be checked within sample limit, use exact computation\n\tif (possibleTriangles <= sampleSize) {\n\t\treturn localClusteringCoefficient(graph, nodeId);\n\t}\n\n\t// Sample pairs uniformly\n\tlet triangleCount = 0;\n\tlet sampled = 0;\n\n\t// Use reservoir sampling style approach for pair selection\n\tfor (let i = 0; i < neighbours.length && sampled < sampleSize; i++) {\n\t\tconst u = neighbours[i];\n\t\tif (u === undefined) continue;\n\n\t\tfor (let j = i + 1; j < neighbours.length && sampled < sampleSize; j++) {\n\t\t\tconst w = neighbours[j];\n\t\t\tif (w === undefined) continue;\n\n\t\t\t// Decide whether to include this pair based on remaining budget\n\t\t\tsampled++;\n\n\t\t\t// Check if u and w are connected\n\t\t\tif (\n\t\t\t\tgraph.getEdge(u, w) !== undefined ||\n\t\t\t\tgraph.getEdge(w, u) !== undefined\n\t\t\t) {\n\t\t\t\ttriangleCount++;\n\t\t\t}\n\t\t}\n\t}\n\n\t// Extrapolate from sample\n\treturn (triangleCount / sampled) * (possibleTriangles / possibleTriangles);\n}\n\n/**\n * Compute clustering coefficients for multiple nodes efficiently.\n *\n * Reuses neighbour sets to avoid repeated iteration.\n *\n * @param graph - The graph to compute on\n * @param nodeIds - The nodes to compute clustering coefficients for\n * @returns Map from nodeId to clustering coefficient\n */\nexport function batchClusteringCoefficients(\n\tgraph: ReadableGraph,\n\tnodeIds: readonly NodeId[],\n): Map<NodeId, number> {\n\tconst results = new Map<NodeId, number>();\n\n\tfor (const nodeId of nodeIds) {\n\t\tresults.set(nodeId, localClusteringCoefficient(graph, nodeId));\n\t}\n\n\treturn results;\n}\n","/**\n * Entropy computation utilities for graph analysis.\n *\n * Shannon entropy measures uncertainty or randomness in a distribution.\n * Used in EDGE and HAE algorithms for heterogeneity-aware expansion.\n *\n * @packageDocumentation\n */\n\n/**\n * Compute Shannon entropy of a probability distribution.\n *\n * Shannon entropy is defined as:\n *   H(X) = -Σ p(x) × log₂(p(x))\n *\n * A uniform distribution has maximum entropy.\n * A deterministic distribution (all probability on one value) has zero entropy.\n *\n * @param probabilities - Array of probabilities (should sum to 1)\n * @returns Entropy in bits (log base 2), or 0 if probabilities are invalid\n */\nexport function shannonEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length === 0) {\n\t\treturn 0;\n\t}\n\n\tlet entropy = 0;\n\tfor (const p of probabilities) {\n\t\t// Skip zero probabilities (log(0) is undefined, but 0 * log(0) = 0)\n\t\tif (p > 0) {\n\t\t\tentropy -= p * Math.log2(p);\n\t\t}\n\t}\n\n\treturn entropy;\n}\n\n/**\n * Compute normalised entropy (entropy divided by maximum possible entropy).\n *\n * Normalised entropy is in [0, 1], where:\n * - 0 means the distribution is deterministic (all mass on one value)\n * - 1 means the distribution is uniform (maximum uncertainty)\n *\n * This is useful for comparing entropy across distributions with different\n * numbers of possible values.\n *\n * @param probabilities - Array of probabilities (should sum to 1)\n * @returns Normalised entropy in [0, 1], or 0 if only one category\n */\nexport function normalisedEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length <= 1) {\n\t\treturn 0;\n\t}\n\n\tconst H = shannonEntropy(probabilities);\n\tconst Hmax = Math.log2(probabilities.length);\n\n\tif (Hmax === 0) {\n\t\treturn 0;\n\t}\n\n\treturn H / Hmax;\n}\n\n/**\n * Compute entropy from a frequency count.\n *\n * Converts counts to probabilities and then computes entropy.\n * This is a convenience function when you have raw counts rather than\n * normalised probabilities.\n *\n * @param counts - Array of frequency counts\n * @returns Entropy in bits\n */\nexport function entropyFromCounts(counts: readonly number[]): number {\n\tif (counts.length === 0) {\n\t\treturn 0;\n\t}\n\n\tconst total = counts.reduce((sum, c) => sum + c, 0);\n\tif (total === 0) {\n\t\treturn 0;\n\t}\n\n\tconst probabilities = counts.map((c) => c / total);\n\treturn shannonEntropy(probabilities);\n}\n\n/**\n * Compute local type entropy for a node's neighbours.\n *\n * This measures the diversity of types among a node's neighbours.\n * High entropy = heterogeneous neighbourhood (diverse types).\n * Low entropy = homogeneous neighbourhood (similar types).\n *\n * @param neighbourTypes - Array of type labels for neighbours\n * @returns Normalised entropy in [0, 1]\n */\nexport function localTypeEntropy(neighbourTypes: readonly string[]): number {\n\tif (neighbourTypes.length <= 1) {\n\t\treturn 0;\n\t}\n\n\t// Count occurrences of each type\n\tconst typeCounts = new Map<string, number>();\n\tfor (const t of neighbourTypes) {\n\t\ttypeCounts.set(t, (typeCounts.get(t) ?? 0) + 1);\n\t}\n\n\t// If all neighbours are the same type, entropy is 0\n\tif (typeCounts.size === 1) {\n\t\treturn 0;\n\t}\n\n\t// Convert to probability array\n\tconst probabilities: number[] = [];\n\tconst total = neighbourTypes.length;\n\tfor (const count of typeCounts.values()) {\n\t\tprobabilities.push(count / total);\n\t}\n\n\treturn normalisedEntropy(probabilities);\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAyBA,SAAgB,2BACf,OACA,QACS;CACT,MAAM,aAAa,CAAC,GAAG,MAAM,WAAW,QAAQ,OAAO,CAAC;CACxD,MAAM,SAAS,WAAW;AAG1B,KAAI,SAAS,EACZ,QAAO;CAIR,IAAI,gBAAgB;AAEpB,MAAK,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;EAC3C,MAAM,IAAI,WAAW;AACrB,MAAI,MAAM,KAAA,EAAW;AAErB,OAAK,IAAI,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;GAC/C,MAAM,IAAI,WAAW;AACrB,OAAI,MAAM,KAAA,EAAW;AAGrB,OACC,MAAM,QAAQ,GAAG,EAAE,KAAK,KAAA,KACxB,MAAM,QAAQ,GAAG,EAAE,KAAK,KAAA,EAExB;;;CAOH,MAAM,oBAAqB,UAAU,SAAS,KAAM;AAEpD,QAAO,gBAAgB;;;;;;;;;;;;;AAcxB,SAAgB,iCACf,OACA,QACA,aAAa,KACJ;CACT,MAAM,aAAa,CAAC,GAAG,MAAM,WAAW,QAAQ,OAAO,CAAC;CACxD,MAAM,SAAS,WAAW;AAE1B,KAAI,SAAS,EACZ,QAAO;CAGR,MAAM,oBAAqB,UAAU,SAAS,KAAM;AAGpD,KAAI,qBAAqB,WACxB,QAAO,2BAA2B,OAAO,OAAO;CAIjD,IAAI,gBAAgB;CACpB,IAAI,UAAU;AAGd,MAAK,IAAI,IAAI,GAAG,IAAI,WAAW,UAAU,UAAU,YAAY,KAAK;EACnE,MAAM,IAAI,WAAW;AACrB,MAAI,MAAM,KAAA,EAAW;AAErB,OAAK,IAAI,IAAI,IAAI,GAAG,IAAI,WAAW,UAAU,UAAU,YAAY,KAAK;GACvE,MAAM,IAAI,WAAW;AACrB,OAAI,MAAM,KAAA,EAAW;AAGrB;AAGA,OACC,MAAM,QAAQ,GAAG,EAAE,KAAK,KAAA,KACxB,MAAM,QAAQ,GAAG,EAAE,KAAK,KAAA,EAExB;;;AAMH,QAAQ,gBAAgB,WAAY,oBAAoB;;;;;;;;;;;AAYzD,SAAgB,4BACf,OACA,SACsB;CACtB,MAAM,0BAAU,IAAI,KAAqB;AAEzC,MAAK,MAAM,UAAU,QACpB,SAAQ,IAAI,QAAQ,2BAA2B,OAAO,OAAO,CAAC;AAG/D,QAAO;;;;;;;;;;;;;;;;;;;;;;;;AC3HR,SAAgB,eAAe,eAA0C;AACxE,KAAI,cAAc,WAAW,EAC5B,QAAO;CAGR,IAAI,UAAU;AACd,MAAK,MAAM,KAAK,cAEf,KAAI,IAAI,EACP,YAAW,IAAI,KAAK,KAAK,EAAE;AAI7B,QAAO;;;;;;;;;;;;;;;AAgBR,SAAgB,kBAAkB,eAA0C;AAC3E,KAAI,cAAc,UAAU,EAC3B,QAAO;CAGR,MAAM,IAAI,eAAe,cAAc;CACvC,MAAM,OAAO,KAAK,KAAK,cAAc,OAAO;AAE5C,KAAI,SAAS,EACZ,QAAO;AAGR,QAAO,IAAI;;;;;;;;;;;;AAaZ,SAAgB,kBAAkB,QAAmC;AACpE,KAAI,OAAO,WAAW,EACrB,QAAO;CAGR,MAAM,QAAQ,OAAO,QAAQ,KAAK,MAAM,MAAM,GAAG,EAAE;AACnD,KAAI,UAAU,EACb,QAAO;AAIR,QAAO,eADe,OAAO,KAAK,MAAM,IAAI,MAAM,CACd;;;;;;;;;;;;AAarC,SAAgB,iBAAiB,gBAA2C;AAC3E,KAAI,eAAe,UAAU,EAC5B,QAAO;CAIR,MAAM,6BAAa,IAAI,KAAqB;AAC5C,MAAK,MAAM,KAAK,eACf,YAAW,IAAI,IAAI,WAAW,IAAI,EAAE,IAAI,KAAK,EAAE;AAIhD,KAAI,WAAW,SAAS,EACvB,QAAO;CAIR,MAAM,gBAA0B,EAAE;CAClC,MAAM,QAAQ,eAAe;AAC7B,MAAK,MAAM,SAAS,WAAW,QAAQ,CACtC,eAAc,KAAK,QAAQ,MAAM;AAGlC,QAAO,kBAAkB,cAAc"}

package/dist/utils/index.d.ts

@@ -3,7 +3,7 @@
  *
  * @packageDocumentation
  */
-export { localClusteringCoefficient, approximateClusteringCoefficient, batchClusteringCoefficients, } from './clustering-coefficient';
-export { miniBatchKMeans, normaliseFeatures, zScoreNormalise, type FeatureVector3D, type LabelledFeature, type KMeansResult, type KMeansOptions, } from './kmeans';
-export { shannonEntropy, normalisedEntropy, entropyFromCounts, localTypeEntropy, } from './entropy';
+export * from './clustering-coefficient';
+export * from './kmeans';
+export * from './entropy';
 //# sourceMappingURL=index.d.ts.map

package/dist/utils/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/utils/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EACN,0BAA0B,EAC1B,gCAAgC,EAChC,2BAA2B,GAC3B,MAAM,0BAA0B,CAAC;AAElC,OAAO,EACN,eAAe,EACf,iBAAiB,EACjB,eAAe,EACf,KAAK,eAAe,EACpB,KAAK,eAAe,EACpB,KAAK,YAAY,EACjB,KAAK,aAAa,GAClB,MAAM,UAAU,CAAC;AAElB,OAAO,EACN,cAAc,EACd,iBAAiB,EACjB,iBAAiB,EACjB,gBAAgB,GAChB,MAAM,WAAW,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/utils/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,cAAc,0BAA0B,CAAC;AACzC,cAAc,UAAU,CAAC;AACzB,cAAc,WAAW,CAAC"}

package/dist/utils/index.js

@@ -1,3 +1,165 @@
-import { n as normaliseFeatures, t as miniBatchKMeans } from "../kmeans-DgbsOznU.js";
-import { a as approximateClusteringCoefficient, i as shannonEntropy, n as localTypeEntropy, o as batchClusteringCoefficients, r as normalisedEntropy, s as localClusteringCoefficient, t as entropyFromCounts } from "../utils-Q_akvlMn.js";
-export { approximateClusteringCoefficient, batchClusteringCoefficients, entropyFromCounts, localClusteringCoefficient, localTypeEntropy, miniBatchKMeans, normaliseFeatures, normaliseFeatures as zScoreNormalise, normalisedEntropy, shannonEntropy };
+import { n as miniBatchKMeans, r as normaliseFeatures, t as _computeMean } from "../kmeans-87ExSUNZ.js";
+//#region src/utils/clustering-coefficient.ts
+/**
+* Compute the local clustering coefficient for a single node.
+*
+* The clustering coefficient is defined as:
+*   CC(v) = (triangles through v) / (possible triangles)
+*   CC(v) = 2 * |{(u,w) : u,w in N(v), (u,w) in E}| / (deg(v) * (deg(v) - 1))
+*
+* For nodes with degree < 2, the clustering coefficient is 0.
+*
+* @param graph - The graph to compute on
+* @param nodeId - The node to compute clustering coefficient for
+* @returns The clustering coefficient in [0, 1], or 0 if undefined
+*/
+function localClusteringCoefficient(graph, nodeId) {
+	const neighbours = [...graph.neighbours(nodeId, "both")];
+	const degree = neighbours.length;
+	if (degree < 2) return 0;
+	let triangleCount = 0;
+	for (let i = 0; i < neighbours.length; i++) {
+		const u = neighbours[i];
+		if (u === void 0) continue;
+		for (let j = i + 1; j < neighbours.length; j++) {
+			const w = neighbours[j];
+			if (w === void 0) continue;
+			if (graph.getEdge(u, w) !== void 0 || graph.getEdge(w, u) !== void 0) triangleCount++;
+		}
+	}
+	const possibleTriangles = degree * (degree - 1) / 2;
+	return triangleCount / possibleTriangles;
+}
+/**
+* Compute approximate local clustering coefficient using sampling.
+*
+* For nodes with many neighbours, this samples neighbour pairs rather than
+* checking all pairs. Useful for large graphs where exact computation is expensive.
+*
+* @param graph - The graph to compute on
+* @param nodeId - The node to compute clustering coefficient for
+* @param sampleSize - Maximum number of neighbour pairs to check (default: 100)
+* @returns The approximate clustering coefficient in [0, 1]
+*/
+function approximateClusteringCoefficient(graph, nodeId, sampleSize = 100) {
+	const neighbours = [...graph.neighbours(nodeId, "both")];
+	const degree = neighbours.length;
+	if (degree < 2) return 0;
+	const possibleTriangles = degree * (degree - 1) / 2;
+	if (possibleTriangles <= sampleSize) return localClusteringCoefficient(graph, nodeId);
+	let triangleCount = 0;
+	let sampled = 0;
+	for (let i = 0; i < neighbours.length && sampled < sampleSize; i++) {
+		const u = neighbours[i];
+		if (u === void 0) continue;
+		for (let j = i + 1; j < neighbours.length && sampled < sampleSize; j++) {
+			const w = neighbours[j];
+			if (w === void 0) continue;
+			sampled++;
+			if (graph.getEdge(u, w) !== void 0 || graph.getEdge(w, u) !== void 0) triangleCount++;
+		}
+	}
+	return triangleCount / sampled * (possibleTriangles / possibleTriangles);
+}
+/**
+* Compute clustering coefficients for multiple nodes efficiently.
+*
+* Reuses neighbour sets to avoid repeated iteration.
+*
+* @param graph - The graph to compute on
+* @param nodeIds - The nodes to compute clustering coefficients for
+* @returns Map from nodeId to clustering coefficient
+*/
+function batchClusteringCoefficients(graph, nodeIds) {
+	const results = /* @__PURE__ */ new Map();
+	for (const nodeId of nodeIds) results.set(nodeId, localClusteringCoefficient(graph, nodeId));
+	return results;
+}
+//#endregion
+//#region src/utils/entropy.ts
+/**
+* Entropy computation utilities for graph analysis.
+*
+* Shannon entropy measures uncertainty or randomness in a distribution.
+* Used in EDGE and HAE algorithms for heterogeneity-aware expansion.
+*
+* @packageDocumentation
+*/
+/**
+* Compute Shannon entropy of a probability distribution.
+*
+* Shannon entropy is defined as:
+*   H(X) = -Σ p(x) × log₂(p(x))
+*
+* A uniform distribution has maximum entropy.
+* A deterministic distribution (all probability on one value) has zero entropy.
+*
+* @param probabilities - Array of probabilities (should sum to 1)
+* @returns Entropy in bits (log base 2), or 0 if probabilities are invalid
+*/
+function shannonEntropy(probabilities) {
+	if (probabilities.length === 0) return 0;
+	let entropy = 0;
+	for (const p of probabilities) if (p > 0) entropy -= p * Math.log2(p);
+	return entropy;
+}
+/**
+* Compute normalised entropy (entropy divided by maximum possible entropy).
+*
+* Normalised entropy is in [0, 1], where:
+* - 0 means the distribution is deterministic (all mass on one value)
+* - 1 means the distribution is uniform (maximum uncertainty)
+*
+* This is useful for comparing entropy across distributions with different
+* numbers of possible values.
+*
+* @param probabilities - Array of probabilities (should sum to 1)
+* @returns Normalised entropy in [0, 1], or 0 if only one category
+*/
+function normalisedEntropy(probabilities) {
+	if (probabilities.length <= 1) return 0;
+	const H = shannonEntropy(probabilities);
+	const Hmax = Math.log2(probabilities.length);
+	if (Hmax === 0) return 0;
+	return H / Hmax;
+}
+/**
+* Compute entropy from a frequency count.
+*
+* Converts counts to probabilities and then computes entropy.
+* This is a convenience function when you have raw counts rather than
+* normalised probabilities.
+*
+* @param counts - Array of frequency counts
+* @returns Entropy in bits
+*/
+function entropyFromCounts(counts) {
+	if (counts.length === 0) return 0;
+	const total = counts.reduce((sum, c) => sum + c, 0);
+	if (total === 0) return 0;
+	return shannonEntropy(counts.map((c) => c / total));
+}
+/**
+* Compute local type entropy for a node's neighbours.
+*
+* This measures the diversity of types among a node's neighbours.
+* High entropy = heterogeneous neighbourhood (diverse types).
+* Low entropy = homogeneous neighbourhood (similar types).
+*
+* @param neighbourTypes - Array of type labels for neighbours
+* @returns Normalised entropy in [0, 1]
+*/
+function localTypeEntropy(neighbourTypes) {
+	if (neighbourTypes.length <= 1) return 0;
+	const typeCounts = /* @__PURE__ */ new Map();
+	for (const t of neighbourTypes) typeCounts.set(t, (typeCounts.get(t) ?? 0) + 1);
+	if (typeCounts.size === 1) return 0;
+	const probabilities = [];
+	const total = neighbourTypes.length;
+	for (const count of typeCounts.values()) probabilities.push(count / total);
+	return normalisedEntropy(probabilities);
+}
+//#endregion
+export { _computeMean, approximateClusteringCoefficient, batchClusteringCoefficients, entropyFromCounts, localClusteringCoefficient, localTypeEntropy, miniBatchKMeans, normaliseFeatures, normaliseFeatures as zScoreNormalise, normalisedEntropy, shannonEntropy };
+
+//# sourceMappingURL=index.js.map

package/dist/utils/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","names":[],"sources":["../../src/utils/clustering-coefficient.ts","../../src/utils/entropy.ts"],"sourcesContent":["/**\n * Clustering coefficient computation for graph nodes.\n *\n * The local clustering coefficient measures how close a node's neighbours\n * are to being a complete graph (clique). It is used in SPAN MI variant\n * and GRASP seed selection.\n *\n * @packageDocumentation\n */\n\nimport type { ReadableGraph, NodeId } from \"../graph\";\n\n/**\n * Compute the local clustering coefficient for a single node.\n *\n * The clustering coefficient is defined as:\n *   CC(v) = (triangles through v) / (possible triangles)\n *   CC(v) = 2 * |{(u,w) : u,w in N(v), (u,w) in E}| / (deg(v) * (deg(v) - 1))\n *\n * For nodes with degree < 2, the clustering coefficient is 0.\n *\n * @param graph - The graph to compute on\n * @param nodeId - The node to compute clustering coefficient for\n * @returns The clustering coefficient in [0, 1], or 0 if undefined\n */\nexport function localClusteringCoefficient(\n\tgraph: ReadableGraph,\n\tnodeId: NodeId,\n): number {\n\tconst neighbours = [...graph.neighbours(nodeId, \"both\")];\n\tconst degree = neighbours.length;\n\n\t// Nodes with degree < 2 have no possible triangles\n\tif (degree < 2) {\n\t\treturn 0;\n\t}\n\n\t// Count actual triangles: pairs of neighbours that are connected\n\tlet triangleCount = 0;\n\n\tfor (let i = 0; i < neighbours.length; i++) {\n\t\tconst u = neighbours[i];\n\t\tif (u === undefined) continue;\n\n\t\tfor (let j = i + 1; j < neighbours.length; j++) {\n\t\t\tconst w = neighbours[j];\n\t\t\tif (w === undefined) continue;\n\n\t\t\t// Check if u and w are connected\n\t\t\tif (\n\t\t\t\tgraph.getEdge(u, w) !== undefined ||\n\t\t\t\tgraph.getEdge(w, u) !== undefined\n\t\t\t) {\n\t\t\t\ttriangleCount++;\n\t\t\t}\n\t\t}\n\t}\n\n\t// Possible triangles: deg * (deg - 1) / 2 pairs\n\t// We multiply by 2 because each triangle is counted once\n\tconst possibleTriangles = (degree * (degree - 1)) / 2;\n\n\treturn triangleCount / possibleTriangles;\n}\n\n/**\n * Compute approximate local clustering coefficient using sampling.\n *\n * For nodes with many neighbours, this samples neighbour pairs rather than\n * checking all pairs. Useful for large graphs where exact computation is expensive.\n *\n * @param graph - The graph to compute on\n * @param nodeId - The node to compute clustering coefficient for\n * @param sampleSize - Maximum number of neighbour pairs to check (default: 100)\n * @returns The approximate clustering coefficient in [0, 1]\n */\nexport function approximateClusteringCoefficient(\n\tgraph: ReadableGraph,\n\tnodeId: NodeId,\n\tsampleSize = 100,\n): number {\n\tconst neighbours = [...graph.neighbours(nodeId, \"both\")];\n\tconst degree = neighbours.length;\n\n\tif (degree < 2) {\n\t\treturn 0;\n\t}\n\n\tconst possibleTriangles = (degree * (degree - 1)) / 2;\n\n\t// If all pairs can be checked within sample limit, use exact computation\n\tif (possibleTriangles <= sampleSize) {\n\t\treturn localClusteringCoefficient(graph, nodeId);\n\t}\n\n\t// Sample pairs uniformly\n\tlet triangleCount = 0;\n\tlet sampled = 0;\n\n\t// Use reservoir sampling style approach for pair selection\n\tfor (let i = 0; i < neighbours.length && sampled < sampleSize; i++) {\n\t\tconst u = neighbours[i];\n\t\tif (u === undefined) continue;\n\n\t\tfor (let j = i + 1; j < neighbours.length && sampled < sampleSize; j++) {\n\t\t\tconst w = neighbours[j];\n\t\t\tif (w === undefined) continue;\n\n\t\t\t// Decide whether to include this pair based on remaining budget\n\t\t\tsampled++;\n\n\t\t\t// Check if u and w are connected\n\t\t\tif (\n\t\t\t\tgraph.getEdge(u, w) !== undefined ||\n\t\t\t\tgraph.getEdge(w, u) !== undefined\n\t\t\t) {\n\t\t\t\ttriangleCount++;\n\t\t\t}\n\t\t}\n\t}\n\n\t// Extrapolate from sample\n\treturn (triangleCount / sampled) * (possibleTriangles / possibleTriangles);\n}\n\n/**\n * Compute clustering coefficients for multiple nodes efficiently.\n *\n * Reuses neighbour sets to avoid repeated iteration.\n *\n * @param graph - The graph to compute on\n * @param nodeIds - The nodes to compute clustering coefficients for\n * @returns Map from nodeId to clustering coefficient\n */\nexport function batchClusteringCoefficients(\n\tgraph: ReadableGraph,\n\tnodeIds: readonly NodeId[],\n): Map<NodeId, number> {\n\tconst results = new Map<NodeId, number>();\n\n\tfor (const nodeId of nodeIds) {\n\t\tresults.set(nodeId, localClusteringCoefficient(graph, nodeId));\n\t}\n\n\treturn results;\n}\n","/**\n * Entropy computation utilities for graph analysis.\n *\n * Shannon entropy measures uncertainty or randomness in a distribution.\n * Used in EDGE and HAE algorithms for heterogeneity-aware expansion.\n *\n * @packageDocumentation\n */\n\n/**\n * Compute Shannon entropy of a probability distribution.\n *\n * Shannon entropy is defined as:\n *   H(X) = -Σ p(x) × log₂(p(x))\n *\n * A uniform distribution has maximum entropy.\n * A deterministic distribution (all probability on one value) has zero entropy.\n *\n * @param probabilities - Array of probabilities (should sum to 1)\n * @returns Entropy in bits (log base 2), or 0 if probabilities are invalid\n */\nexport function shannonEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length === 0) {\n\t\treturn 0;\n\t}\n\n\tlet entropy = 0;\n\tfor (const p of probabilities) {\n\t\t// Skip zero probabilities (log(0) is undefined, but 0 * log(0) = 0)\n\t\tif (p > 0) {\n\t\t\tentropy -= p * Math.log2(p);\n\t\t}\n\t}\n\n\treturn entropy;\n}\n\n/**\n * Compute normalised entropy (entropy divided by maximum possible entropy).\n *\n * Normalised entropy is in [0, 1], where:\n * - 0 means the distribution is deterministic (all mass on one value)\n * - 1 means the distribution is uniform (maximum uncertainty)\n *\n * This is useful for comparing entropy across distributions with different\n * numbers of possible values.\n *\n * @param probabilities - Array of probabilities (should sum to 1)\n * @returns Normalised entropy in [0, 1], or 0 if only one category\n */\nexport function normalisedEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length <= 1) {\n\t\treturn 0;\n\t}\n\n\tconst H = shannonEntropy(probabilities);\n\tconst Hmax = Math.log2(probabilities.length);\n\n\tif (Hmax === 0) {\n\t\treturn 0;\n\t}\n\n\treturn H / Hmax;\n}\n\n/**\n * Compute entropy from a frequency count.\n *\n * Converts counts to probabilities and then computes entropy.\n * This is a convenience function when you have raw counts rather than\n * normalised probabilities.\n *\n * @param counts - Array of frequency counts\n * @returns Entropy in bits\n */\nexport function entropyFromCounts(counts: readonly number[]): number {\n\tif (counts.length === 0) {\n\t\treturn 0;\n\t}\n\n\tconst total = counts.reduce((sum, c) => sum + c, 0);\n\tif (total === 0) {\n\t\treturn 0;\n\t}\n\n\tconst probabilities = counts.map((c) => c / total);\n\treturn shannonEntropy(probabilities);\n}\n\n/**\n * Compute local type entropy for a node's neighbours.\n *\n * This measures the diversity of types among a node's neighbours.\n * High entropy = heterogeneous neighbourhood (diverse types).\n * Low entropy = homogeneous neighbourhood (similar types).\n *\n * @param neighbourTypes - Array of type labels for neighbours\n * @returns Normalised entropy in [0, 1]\n */\nexport function localTypeEntropy(neighbourTypes: readonly string[]): number {\n\tif (neighbourTypes.length <= 1) {\n\t\treturn 0;\n\t}\n\n\t// Count occurrences of each type\n\tconst typeCounts = new Map<string, number>();\n\tfor (const t of neighbourTypes) {\n\t\ttypeCounts.set(t, (typeCounts.get(t) ?? 0) + 1);\n\t}\n\n\t// If all neighbours are the same type, entropy is 0\n\tif (typeCounts.size === 1) {\n\t\treturn 0;\n\t}\n\n\t// Convert to probability array\n\tconst probabilities: number[] = [];\n\tconst total = neighbourTypes.length;\n\tfor (const count of typeCounts.values()) {\n\t\tprobabilities.push(count / total);\n\t}\n\n\treturn normalisedEntropy(probabilities);\n}\n"],"mappings":";;;;;;;;;;;;;;;AAyBA,SAAgB,2BACf,OACA,QACS;CACT,MAAM,aAAa,CAAC,GAAG,MAAM,WAAW,QAAQ,OAAO,CAAC;CACxD,MAAM,SAAS,WAAW;AAG1B,KAAI,SAAS,EACZ,QAAO;CAIR,IAAI,gBAAgB;AAEpB,MAAK,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;EAC3C,MAAM,IAAI,WAAW;AACrB,MAAI,MAAM,KAAA,EAAW;AAErB,OAAK,IAAI,IAAI,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;GAC/C,MAAM,IAAI,WAAW;AACrB,OAAI,MAAM,KAAA,EAAW;AAGrB,OACC,MAAM,QAAQ,GAAG,EAAE,KAAK,KAAA,KACxB,MAAM,QAAQ,GAAG,EAAE,KAAK,KAAA,EAExB;;;CAOH,MAAM,oBAAqB,UAAU,SAAS,KAAM;AAEpD,QAAO,gBAAgB;;;;;;;;;;;;;AAcxB,SAAgB,iCACf,OACA,QACA,aAAa,KACJ;CACT,MAAM,aAAa,CAAC,GAAG,MAAM,WAAW,QAAQ,OAAO,CAAC;CACxD,MAAM,SAAS,WAAW;AAE1B,KAAI,SAAS,EACZ,QAAO;CAGR,MAAM,oBAAqB,UAAU,SAAS,KAAM;AAGpD,KAAI,qBAAqB,WACxB,QAAO,2BAA2B,OAAO,OAAO;CAIjD,IAAI,gBAAgB;CACpB,IAAI,UAAU;AAGd,MAAK,IAAI,IAAI,GAAG,IAAI,WAAW,UAAU,UAAU,YAAY,KAAK;EACnE,MAAM,IAAI,WAAW;AACrB,MAAI,MAAM,KAAA,EAAW;AAErB,OAAK,IAAI,IAAI,IAAI,GAAG,IAAI,WAAW,UAAU,UAAU,YAAY,KAAK;GACvE,MAAM,IAAI,WAAW;AACrB,OAAI,MAAM,KAAA,EAAW;AAGrB;AAGA,OACC,MAAM,QAAQ,GAAG,EAAE,KAAK,KAAA,KACxB,MAAM,QAAQ,GAAG,EAAE,KAAK,KAAA,EAExB;;;AAMH,QAAQ,gBAAgB,WAAY,oBAAoB;;;;;;;;;;;AAYzD,SAAgB,4BACf,OACA,SACsB;CACtB,MAAM,0BAAU,IAAI,KAAqB;AAEzC,MAAK,MAAM,UAAU,QACpB,SAAQ,IAAI,QAAQ,2BAA2B,OAAO,OAAO,CAAC;AAG/D,QAAO;;;;;;;;;;;;;;;;;;;;;;;;AC3HR,SAAgB,eAAe,eAA0C;AACxE,KAAI,cAAc,WAAW,EAC5B,QAAO;CAGR,IAAI,UAAU;AACd,MAAK,MAAM,KAAK,cAEf,KAAI,IAAI,EACP,YAAW,IAAI,KAAK,KAAK,EAAE;AAI7B,QAAO;;;;;;;;;;;;;;;AAgBR,SAAgB,kBAAkB,eAA0C;AAC3E,KAAI,cAAc,UAAU,EAC3B,QAAO;CAGR,MAAM,IAAI,eAAe,cAAc;CACvC,MAAM,OAAO,KAAK,KAAK,cAAc,OAAO;AAE5C,KAAI,SAAS,EACZ,QAAO;AAGR,QAAO,IAAI;;;;;;;;;;;;AAaZ,SAAgB,kBAAkB,QAAmC;AACpE,KAAI,OAAO,WAAW,EACrB,QAAO;CAGR,MAAM,QAAQ,OAAO,QAAQ,KAAK,MAAM,MAAM,GAAG,EAAE;AACnD,KAAI,UAAU,EACb,QAAO;AAIR,QAAO,eADe,OAAO,KAAK,MAAM,IAAI,MAAM,CACd;;;;;;;;;;;;AAarC,SAAgB,iBAAiB,gBAA2C;AAC3E,KAAI,eAAe,UAAU,EAC5B,QAAO;CAIR,MAAM,6BAAa,IAAI,KAAqB;AAC5C,MAAK,MAAM,KAAK,eACf,YAAW,IAAI,IAAI,WAAW,IAAI,EAAE,IAAI,KAAK,EAAE;AAIhD,KAAI,WAAW,SAAS,EACvB,QAAO;CAIR,MAAM,gBAA0B,EAAE;CAClC,MAAM,QAAQ,eAAe;AAC7B,MAAK,MAAM,SAAS,WAAW,QAAQ,CACtC,eAAc,KAAK,QAAQ,MAAM;AAGlC,QAAO,kBAAkB,cAAc"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "graphwise",
-  "version": "1.1.1",
+  "version": "1.2.0",
   "description": "Graph algorithms for citation network analysis - novel expansion, MI variants, and path ranking",
   "type": "module",
   "sideEffects": false,